Background: Introduction to R (Penn State)

Structure

• 1 credit (i.e. one 50 min meeting per week; semester length)
• Reading quiz before class each week
  
• Group mini-project began in class each week
• Homework set after class each week
• Final project at end of semester

Students

• 33 students (3 dropped)
• most did have a STAT course of some kind (including AP)
• 19 had never used R (another 12 had used it “a little”)
• Half were not (nor planned to) major in statistics or data science

Thoughts before, during, and after course

• heavy use of dplyr, ggplot2, and other tidyverse packages (little base R)
• terminology (e.g., wrangling, glyph-ready, tidy, pipe)
• included topics completely new to Matt (e.g., scraping HTML tables, leaflets)

Background: Data Computing (Macalester)

Tools: Working code

• Provide students with concise and consistent framework of commands (& templates)
• Students generally start with working code that they can modify
• Scaffold with interactive tools
  • Inspecting data (View())
  • Visualizations (xyzGraphBuilder())
• interactive commands produce working code that students can then analyze and modify

Tools: RMarkdown

• advocates common template
  • HTML with embedded Rmd
  • promotes reproducible research habits
  • Students began using it in other STAT classes, and faculty loved it
• prevents excessive dependence on interactive commands
• other document types accessible (LaTeX, slides, etc)

Tools: Other Resources

• RStudio Cheat Sheets
• DataCamp (full access mid-semester)
• Required weekly ask & answer on Piazza >> class stack exchange

Sample Activities

• Take a few minutes to work through a few examples that illustrate the sort of things students might do in a course like this without prerequisites…
  
• You will need a laptop connected to the Internet with a current web browser installed
  
• Feel free to work in small groups

https://mdbeckman.github.io/USCOTS2017Breakout/

PSU Final Project

Instructions were intentionally somewhat vague:

• Reproducible work (i.e. HTML with embedded .Rmd that runs cold)
• Use real data (not loaded from an R package)
• Data wrangling proficiency
• Visualization proficiency
• High quality code, narrative, and overall product

2013 FBI Crime Reporting

• loaded data from FBI website
• basic data wrangling operations: select(), group_by(), summarise(), mutate()
• USMap() choropleths for comparisons among states

Movies in the 21st Century

• scrapes HTML table from www.the-numbers.com
• basic data wrangling operations: filter(), group_by(), summarise(), mutate()
• ggplot() graphics exploring relationships among ticket sales, genre, year, studio, etc
• utilized faceting in several places to invite visual comparisons between groups (e.g. MPAA rating)

FIFA World Rankings Analysis

• investigate trends among confederations and teams over last 10 years
• scrapes HTML table from FIFA World Ranking Website
  
• wrote while loop to utilize an index in the URL to scrape monthly data tables
• join operation adds each monthly table to the master
• reshape data from wide to narrow using gather()
• basic data wrangling operations: select(), filter(), group_by(), summarise()
• ggplot() graphics
• submitted HTML document with embedded .Rmd

Stanley Cup Winners

• investigating trends and commonalities among Stanley Cup champions
• submitted both LaTeX & HTML with embedded .Rmd
• CSV data sourced from Kaggle
• constructed list containing data frames for champions from 1970 to 2012
• used a for loop to iterate over list then create and store ggplot() graphic for each champion in a new list
• used party::ctree() to build competing models based on recursive partitioning
• likelihood-based model comparison

Vegetarian Restaurant Analysis

• exploring trends among restaurants that advertise vegeterian and vegan fare
• basic data wrangling operations: select(), filter(), group_by(), summarise(), mutate()
• interactive map (i.e. leaflet) shows locations of such restaurants throughout the U.S.

Student Code

data1<-unique(data) 
data2<-na.omit(data1)

RestaurantMap <-
 leaflet(data2) %>%
 addTiles() %>%
 addCircleMarkers(radius = 2, color = "red") %>%
 setView( lng =-73.935242, lat =40.730610, zoom = 12)  #New York

RestaurantMap

Vegetarian Restaurant Analysis (Static)

• exploring trends among restaurants that advertise vegeterian and vegan fare
• basic data wrangling operations: select(), group_by(), summarise(), mutate()
• interactive map (i.e. leaflet) shows locations of such restaurants throughout the U.S.

Other interesting projects

History of Reddit

• scrape Reddit data from web
• analysis of new threads over time as well as top Subreddits and SubredditSubs
• used regular expressions to clean text data and identify Subreddits and SubredditSubs

Leading Causes of Death in NYC

• investigation of leading causes of death in NYC
• investigated Alziemers as well citing familial significance

Analysis of Thanksgiving

• investigation of survey data about Thanksgiving traditions
• international student studying cultural norms

Student Outcomes: Core Skills

• data wrangling skills
• data visualization tools
• ability to find answers for themselves

Student Outcomes: Broad Exposure

• Data operations
  • Join tables (left, inner, full, anti)
  • Reshape (convert from wide to narrow and back)
  • Work with dates
• Read data from various sources:
  • HTML tables (e.g. Wikipedia)
  • Yahoo Finance
  • CSV (web & local)
• Regular expressions
  • identify and/or replace a pattern in a character string
  • extracting patterns
• Basic machine learning
• more graphs
  • incorporate 3rd or 4th variable
  • interactive maps (i.e. leaflets)
  • choropleth maps & leaflets
• Smoothing, confidence bands, etc
• Find their own answers
  • R help
  • published “Cheat Sheets”
  • web forums

Where to go from here?

Feel free to revisit some of your initial thoughts about “authentic data science”

• What should be covered by a meaningful first course in data science?
  
• What ambitions do you have for students taking a first data-science course?
  
• In what ways should an authentic data-science course differ from intro stats?

…“without prerequisites”

• What reasonable expectations about students’ abilities and skills coming into the course satisfy “without prerequisites” in the context of your students?
• What computing skills are faithful to the idea of “without prerequisites?”

…or maybe consider some new questions

• What are pros/cons of this sort of course?
  
• Should such a course have a place within a statistics program (major, minor, concentration, etc)?